Abstract Wind turbine manufacturing requires the assembly of large mechanical components, which is crucial to inspect along the production line in order to prevent high reparation costs afterwards. A critical component in this process is the turbine hub, which supports the wind blades and ball bearings allowing the pitch motion. At present, hub inspection is a manual task, which requires expert operators and long... Wind turbine manufacturing requires the assembly of large mechanical components, which is crucial to inspect along the production line in order to prevent high reparation costs afterwards. A critical component in this process is the turbine hub, which supports the wind blades and ball bearings allowing the pitch motion. At present, hub inspection is a manual task, which requires expert operators and long execution time. This paper proposes a novel methodology for the selfadaptive inspection of wind turbine hubs via industrial robots: a set of Critical-To-Quality parameters (CTQs), are inferred from the CAD drawing of wind turbine hub; registration between robot and hub is performed; finally a CAD2robot trajectories planning is accomplished. Methodology is implemented through a Matlab and Simulink Programming Language and combined with an Industrial PC-based control technology Beckhoff TwinCAT 3. Tests with an Fanuc Industrial M-6iB robot arm and R-30iA controller have been successfully performed on re-scaled model of the hub. The flexibility of this methodology allows applications on other industrial contexts, which can benefit from automation. Read More Read Less

Abstract A fuzzy-logic controller supporting the manufacturing of wind turbines and the bolt tightening of their hubs has been designed. The controller embeds assembly error recognition capability and detects tightening faults like misalignment, different threads, cross threads and wrong or small nuts. According to this capability, K-nearest classifiers have been implemented to cluster the output controllers into the diverse fault scenarios. Classifiers make use... A fuzzy-logic controller supporting the manufacturing of wind turbines and the bolt tightening of their hubs has been designed. The controller embeds assembly error recognition capability and detects tightening faults like misalignment, different threads, cross threads and wrong or small nuts. According to this capability, K-nearest classifiers have been implemented to cluster the output controllers into the diverse fault scenarios. Classifiers make use of the time of execution of the tightening process, the final angular position of and applied torque of the tightening tool, the resultant clamping force and possible combinations of those parameters. Two classes and five classes configurations are considered: classifiers are initially asked to discriminate between fault and no fault scenarios (e.g. two classes); then, five classes are considered according to five different fault situations (i.e. regular tightening, bolt misalignment, dissimilar threads of bolt and nut, missing nut and small bolt). Classifiers performances are estimated in terms of resubstitution and cross-validation loss. Confusion matrixes of actual and predicted classification are also evaluated for each classifier. The low computational cost of the proposed classifiers suggests directly implementing these algorithms on micro-controller and physical computing, which may be straight integrated within the tightening tool. Read More Read Less

Abstract The sparse representation based classifier (SRC) is a classical representation method for classification. The solution of SRC is obtained by l1 norm minimization, which can not obtain the closed form solution. Thus, the computational complexity of SRC is a little high. The collaborative representation classifier (CRC) is another classical method for classification. The solution of CRC is obtained by l2 norm minimization, from... The sparse representation based classifier (SRC) is a classical representation method for classification. The solution of SRC is obtained by l1 norm minimization, which can not obtain the closed form solution. Thus, the computational complexity of SRC is a little high. The collaborative representation classifier (CRC) is another classical method for classification. The solution of CRC is obtained by l2 norm minimization, from the l2 norm minimization, it can obtain the closed form solution, which makes the computational complexity of CRC is much lower than SRC. Although CRC is effective for classification, there are also some problems about CRC. Under some conditions, some test samples may be misclassified by CRC. This paper proposes a local CRC method, which is called KNN-CRC. This method firstly chooses K nearest neighbors of a test sample from all the training samples, then given a test sample, the test sample is represented by these K training samples. The solution of KNN-CRC is obtained by l2 norm minimization, and the size of K is much smaller than the total number of all training samples. Thus, the computational complexity of KNN-CRC is much lower than SRC and CRC. Furthermore, the extensive experiments show that the proposed KNN-CRC can obtain very competitive classification results compared with other methods. Read More Read Less